NEEW, May 2018

Context

Evidence in experimental and behavioral economics suggests the following:

  • people have a preference for honesty (or against guilt or lying)
    • see Gneezy (2005), Gneezy et al (2013), Dreber & Johannssen (2010), Kartik (2009), and many others
  • people have a utility function with standard properties
    • they have social preferences because they dislike lying
    • lying imposes a utility cost
    • extensive margin: cost to lying at all
    • intensive margin: cost in size of lie

Lying in the real world?

But there's lots of evidence of people lying for profit

  • people lied about MBS and CDOs prior to the financial crisis
  • financial advisers who are not fiduciaries
  • medical professionals who benefit from prescribing certain drugs
  • lawyers displaying prosecutorial dishonesty
  • current political climate?
    • do politicians who lie crowd out those who are unwilling (or are as unwilling) to do so?
    • what does this do to the market for politicans?

Motivating questions & gaps in literature

  • If honesty is preference-based, how vulnerable are the preferences to institutional setting?
    • Do preferences override setting, or are preferences context-dependent?
  • How do people respond to competition?
  • Can competition encourage/discourage full deception rather than merely 'disguised' lies (FFH, 2013)?
  • How do outside offers affect choices to compete & to lie?
    • can providing better remuneration ensure honesty?
  • Do people select into (more) honest settings?
  • Do people adapt to deception and lie more? (Garrett et al, 2016)

Design Context

  • Dice-rolling tasks
  • Fischbacher & Föllmi-Heusi (2013)
  • Dannenberg & Kachatryan (WP)
  • Pay $1 per 'pip' of the die that is reported, pay $0 for a 6.
  • Among subjects who participated more than once, higher rolls reported.
  • D & K introduce "group decisions" & competition, but don't observe more deception.

Innovations

We innovate across two main components:

  1. Providing an outside option task for types with strong lying aversion. By varying the expected payoff from this task, we are able to affect the choices of marginal subjects.
  2. Repeating the task within small groups.

Theory & Hypotheses

  • People varying in guilt aversion & lying
    • Captured by lying cost, \(C_i\)
  • The composition of groups affects my choice to enter or lie
    • belief fraction of others are honest = \(\lambda\)
    • & belief that fraction \(1 - \lambda\) are dishonest
  • Measure margin of entry into competitive task
  • Measure margin of choices on average

Theory: Amount of Deception

Deception and Honesty

Theory: Costs of Lying & Beliefs

Deception and Honesty

Theory & Hypotheses: Basics

  • H1: More enter the competitive task when the outside option is low

  • H2: More enter the competitive task when others are honest

Theory & Hypotheses: Types

  • Type H1: Conditionally honest types react with initial honesty that decreases over time, where they remain in the competitive task.

  • Type H2: Unconditionally honest types react by either remaining in the task (possible for honest groups) or moving to the uncompetitive task.

Theory & Hypotheses: Equilibrium

  • Equilibrium H1: In Group-High, either all group members or no group members are honest

  • Equilibrium H2: In Group-Low, either all group members are dishonest, or at most two group members

Our Design: Competition with outside offers

  • Groups of 6 subjects
  • Choice of two tasks, A or B.
  • Task A – Outside Offer
  • Task B – Deception Task

Our Design: Competition with outside offers

Task A (Outside Offer):

  • If A is chosen, subject reports a 10-sided die roll, which is compared to balls in an urn:
    • General idea: n even balls, n odd ball, 5 any, 95 - 2n neither
    • High Treatment: 40 even balls, 40 odd balls, 5 any, 15 neither balls.
    • Low Treatment: 20 even balls, 20 odd balls, 5 any, 55 neither balls.
    • If die report matches odd/even/any get paid $15.
    • Otherwise payment is $5.00.

Baseline: Competition with high outside offer

Task B (Deception Task):

  • If B is chosen, the subject reports a 10-sided die roll.
    • Roll compared with another player's reported 10-side die roll.
    • If higher, player obtains $15; if lower, then $5; 50-50 on a tie.

Our design (HLW, contd.)

Individual Treatments (High and Low) Same choice between Task A and Task B.

Task A

  • High and Low treatments as defined previously

Task B

  • If B is chosen, the subject reports a 10-sided die roll.
    • Roll compared with a computer-rolled 10-side die with each number equally likely to be chosen.
    • If higher, player obtains $15, if lower $5; 50-50 chance on a tie.

Our design (HLW, contd.)

Additional details

  • Subject repeats their choice 30 times.
  • We observe the margin of choosing Tasks A or B (extensive)
  • On choosing B, we measure the distribution of reports (intensive because repeated within subject)
  • Contrast with theoretical distribution and across treatments

Design Comparison

Treatment Source of Y Fixed Prob. Sessions Groups Subjects Decisions
Indiv-L Uniform Draw \(\frac{25}{100}\) 1 24 24 720
Indiv-H Uniform Draw \(\frac{45}{100}\) 1 24 24 720
Group-L Endogenous \(\frac{25}{100}\) 3 18 72 2,160
Group-H Endogenous \(\frac{45}{100}\) 3 18 72 2,160
Total - - 8 84 192 5,760

Preliminary Results

  • Competition increases deception
  • Offering honest people a decent outside offer reduces their selection into the competitive task
  • With low outside offers, more people deceive on the extensive, but deception on the intensive margin is lower
    • honest types don't want to deceive quite as much
  • Less deception in the decision environment

Average rolls by choice

Rolls by Treatment

Scatterplot By Treatment

Histogram by treatment

Tasks by Group and Distance from Uniform, 1st Half

Tasks by Group and Distance from Uniform, 2nd Half

Individual with Task A

Compare Entry Rates & Rolls

  • Compare entry rates into task B at the start and end of the game
  • Examine average rolls at the level of the individual in the first & last five rounds of the game.

Entry Rates: Beginning and End

Histograms of the entry rates

Average Rolls: First 5

Consider a histogram of the individual average roll for the last five rounds by treatment in Task B.

Average Rolls: Last Five

Consider a histogram of the individual average roll for the last five rounds by treatment in Task B.

Plots by treatment over all rounds

Conclusions and contributions?

  • Measure margins not previously measured
  • Repeated choices \(\Rightarrow\) learning
    • others have been one-shot
  • Task is short with a high payoff (incentive compatible)
  • Extensions
    • experiment to examine demand and supply of deception
    • what happens if hiring decisions are based on report? \(\uparrow\) or \(\downarrow\) deception?

Further treatments: Deceptive Computer

T2: Treament "9/10 High Roll" (Decision problem)

  • Same as T1, except computer has a die with 9 faces = 9 and 1 = 0.
  • 50-50 chance of prize if reporting 9.
  • Why? Upper-bound limit on those who want the 50-50 lottery if they are guaranteed to lie.
  • How do we constrain, if at all, deceptive types?

Further Detail on Context

Fischbacher & Föllmi-Heushi (JEEA, 2013)

  • dice-throwing game to measure aggregate deception.
  • subjects roll a 6-sided die and report roll
  • Roll 6 \(\Rightarrow\) CHF 0
  • Roll \(1 \rightarrow 5 \Rightarrow\) equivalent CHF
  • Honesty \(\Rightarrow\) with large \(n\), should be a uniform distribution \(U \sim [0, 5]\)

For example…

FFH's Results

FFH Continued

  • also a "high-stakes" treatment
    • subjects were paid CHF3 per pip reported up to CHF15
  • all one-shot
  • Experiment run after others (add-on)
  • Repeat ubjects more likely to report a higher number the second time.
    • do we 'learn to lie'?

Extension: D & K (2017)

Dannenberg and Khachatryan (2017, working paper)

  • replicate FFH + "competition" + group decisions
  • one-shot competitive choice
    • 1 vs. 1, or
    • in team of 3 vs. team of 3, or
    • 1 vs team of 3

Extension: D & K (2017)

  • 1 vs. 1: subjects no more dishonest than subjects who do not compete
  • 3 vs. 3: more likely to report higher numbers (are more dishonest)
  • Pace: Charness & Sutter (2013) – groups make "more self-interested" decisions

Gaps left by FFH, D & K?

  • Subjects in a "competitive" treatment have no choice about competing
  • Group vs. individual not that interesting to us
  • D & K have no opportunity to "learn to lie"
  • If a subject cannot choose not to compete, is it really a choice to compete?
  • What is the role of an outside offer/opportunity cost?
  • What about the margins of:
    • choosing to compete and
    • reports of those who compete conditional on having chosen to compete?

Potential behaviors (randomized n = 100 rolls)

Preliminary Regression Analysis: Panel Model

For individual \(i\) in group \(g\) and period \(t\): \[P[\mbox{Choose B = 1}]_{igt} = \beta_0 + \beta_1\cdot T_{ig} + \beta_2 Trend + \epsilon_{igt}\] Regressions include the treatment dummies and, in the second specification, trend dummies.

Regression Results

Pr. Chose Competitive Task (Task B)
Basic With Trend
SessionTypeIndividual High 0.33*** 0.33***
(0.02) (0.02)
SessionTypeGroup Low 0.11*** 0.11***
(0.02) (0.02)
SessionTypeGroup High -0.13*** -0.13***
(0.02) (0.02)
Constant 0.54*** 0.49***
(0.02) (0.03)
Observations 5,760 5,760
Adjusted R2 0.10 0.10
Notes: Panel LPM.
***Significant at the 1 percent level

Basic Summary Statistics

Variable Mean/Proportion Std. Dev
Average Roll 5.981 2.939
Proportion Competitive 0.57 0.495
Average Roll For Task A 4.611 2.785
Average Roll for Task B 7.015 2.612